Space-Time Independent Component Analysis of Brain Signals: Component Selection and the Curse of Dimensionality.

Performing Independent Component Analysis (ICA) on biomedical signals is quite commonplace. ICA is usually applied to multi-channel data however not always with great success. In previous work we realized an innovation to standard ICA which we call space-time ICA (ST-ICA). This method brings into play both spatial and temporal/spectral information to perform very powerful extractions and overcomes the individual limitations of ensemble (spatial) ICA and single-channel (temporal) ICA. The cost in implementing ST-ICA is the curse of dimensionality since spatio-temporal analysis of multi-channel physiological data recorded at suitable sampling speeds results in large unwieldy datasets which become impossible to parse without any form of truncation or at least an automated component selection process. Here we address the component selection problem on the application of ST -ICA to real-world neurophysiological data-specifically in extracting seizure data from EEG recordings. We assess the information held in each of the spatio-temporal features resulting from ST-ICA and comment on the development of an efficient method to extract them, as well as using dimensional reduction techniques to reduce the curse of dimensionality resulting successful separation of meaningful physiological data from noisy, artifact laden datasets. Clinical Relevance-These methods will allow for the automatic identification and extraction of poorly defined episodes of physiologically meaningful activity in noisy multi-channel recordings of brain signals.

Introducing a Deflationary Approach to Space-Time ICA that uses temporal methods in Brain Signals Processing.

For the extraction of underlying sources of brain activity, time structure-based techniques for applying Independent Component Analysis (ICA) have been demonstrably more robust than state-of-the-art statistical-based methods, such as FastICA. Since the early application of conventional ICA on electroencephalogram (EEG) recordings, Space-Time ICA (ST-ICA) has emerged as more capable approach for extracting complex underlying activity, but not without the 'curse of dimensionality'. The challenges in the future development of ST-ICA will require a focus on the optimisation of the mixing matrix, and on component clustering techniques. This paper proposes a new optimisation approach for the mixing matrix, which makes ST-ICA more tractable, when using a time structure-based ICA technique, LSDIAG. Such techniques rely on constructing a multi-layer covariance matrix, Cxk of the original dataset to generate the inverse of the mixing matrix; Csk = WCxkWT. This means a simple truncation of the mixing matrix is not appropriate. To overcome this, we propose a deflationary approach to optimise a much smaller mixing matrix - based on the absolute values of the diagonals of the co-variance matrix, Csk, to represent the underlying sources. The preliminary results of the new technique applied to different channels of EEG recorded using the standard 10-20 system - including the full selection of all channels - are very promising.Clinical Relevance-The potential of this deflationary approach for Space-Time ICA, seeks to allow clinicians to identify underlying sources in the brain - that both spatially and spectrally overlap - to be identified, whilst making the 'dimensionality' challenges more tractable. In the long run, applications of this technique could enhance certain brain-computer interface paradigms.

Validation of a Commercial Android Smartwatch as an Activity Monitoring Platform.

Activity monitoring (AM) is an established technique for the assessment of a person's physical activity. With the rapid rise of smartwatch technology, this platform presents an interesting opportunity to use a device for AM that has both the ability to monitor activity and also the ability to interface seamlessly with other healthcare systems. There are questions however around the suitability of smartwatches as monitoring devices. This paper presents a validation of one smartwatch, the ZGPAX S8, for use as an activity monitor. Two experiments are presented: a physical manipulation test and a co-location test. In the physical manipulation test, three S8s are compared to a reference accelerometer under human physical manipulation. In the co-location test, the smartwatch is used alongside a reference device for a period of three hours by four participants to assess both the accelerometer data and the results of processing on data from both devices. Findings from these experiments show that the S8 accelerometer has a good correlation and limits of agreement in the physical manipulation test (r2 ∼ 0.95, CR ∼ 2.5 m/s 2), and excellent correlation and limits of agreement in the analysis of processed data from the co-location experiment (r2 ∼ 0.99, CR ∼ 0.23). From these results, the S8 is evaluated to be a suitable device for AM. Some specific limitations in the S8 are identified such as data range clipping, time drift and sample rate consistency, but these are not found to impact on the suitability of the device once algorithmic processing is applied to the data.

Does feedback on daily activity level from a Smart watch during inpatient stroke rehabilitation increase physical activity levels? Study protocol for a randomized controlled trial.

BACKGROUND: Practicing activities improves recovery after stroke, but many people in hospital do little activity. Feedback on activity using an accelerometer is a potential method to increase activity in hospital inpatients. This study's goal is to investigate the effect of feedback, enabled by a Smart watch, on daily physical activity levels during inpatient stroke rehabilitation and the short-term effects on simple functional activities, primarily mobility. METHODS/DESIGN: A randomized controlled trial will be undertaken within the stroke rehabilitation wards of the Second Affiliated hospital of Anhui University of Traditional Chinese Medicine, Hefei, China. The study participants will be stroke survivors who meet inclusion criteria for the study, primarily: able to participate, no more than 4 months after stroke and walking independently before stroke. Participants will all receive standard local rehabilitation and will be randomly assigned either to receive regular feedback about activity levels, relative to a daily goal tailored by the smart watch over five time periods throughout a working day, or to no feedback, but still wearing the Smart watch. The intervention will last up to 3 weeks, ending sooner if discharged. The data to be collected in all participants include measures of daily activity (Smart watch measure); mobility (Rivermead Mobility Index and 10-metre walking time); independence in personal care (Barthel Activities of Daily Living (ADL) Index); overall activities (the World Health Organization (WHO) Disability Assessment Scale, 12-item version); and quality of life (the Euro-Qol 5L5D). Data will be collected by assessors blinded to allocation of the intervention at baseline, 3 weeks or at discharge (whichever is the sooner); and a reduced data set will be collected at 12 weeks by telephone interview. The primary outcome will be change in daily accelerometer activity scores. Secondary outcomes are compliance and adherence to wearing the watch, and changes in mobility, independence in personal care activities, and health-related quality of life. DISCUSSION: This project is being implemented in a large city hospital with limited resources and limited research experience. There has been a pilot feasibility study using the Smart watch, which highlighted some areas needing change and these are incorporated in this protocol. TRIAL REGISTRATION: ClinicalTrials.gov, NCT02587585 . Registered on 30 September 2015. Chinese Clinical Trial Registry, ChiCTR-IOR-15007179 . Registered on 8 August 2015.

Reconstructing physical activity monitoring outcome measures using an interval sampling approach.

Smart-watches and wearables are becoming more and more popular and this popularity is resulting in an increased interest in their use in research and clinical settings for monitoring physical activity. The battery life of these devices is of some concern, especially when they are put under increased duress through the need to constantly sample from their accelerometers. In this paper a novel approach to sampling from the accelerometer is explored whereby the accelerometer is sampled in intervals to avoid constant sampling. Results show that through reducing the sampling time and sampling in discontinuous intervals that outcome measures of time spent in sedentary, moderate and vigorous activity can be reconstructed to within acceptable error levels when compared to the same outcome measures calculated from continually sampled data.

On the development of a low-cost EMG switch for communication using minimal muscle contractions.

In the final stages of muscle-wasting diseases like Amyotrophic Lateral Sclerosis and in extreme cases of disability the control of muscles is almost completely lost. However, in some cases there can be almost imperceptible residual control of some muscles. This paper presents the results of a study using an affordable, lightweight, portable EMG switch that has been designed in-house for communication in severely disabled or locked-in patients. It is possible to affect communication with the outside world by using these small residual muscle movements; in this case, more specifically muscles around the eyebrow. The subject makes the device "switch", obtaining a binary "yes/no" answer, through imperceptible muscle contractions. In these preliminary proof-of-principle tests we show an accuracy of around 87% when tested on 19 healthy volunteers over a number of measurement protocols.

AART-BC: a sensor system for monitoring Assistive Technology use beyond the clinic.

A wide range of assistive and rehabilitative technologies (ART) are available to assist with mobility and upper limb function. However, anecdotal evidence suggests many of the devices prescribed, or purchased, are either poorly used, or rejected entirely. This situation is costly, both for the healthcare provider and the user, and may be leading to secondary consequences, such as falls and/or social isolation. This paper reports on the development and initial feasibility testing of a system for monitoring when and how assistive devices are used outside of the clinic setting, and feeding this information to the device user themselves and/or prescribing clinician (where appropriate). Illustrative data from multiple time-synchronized device and body worn sensors are presented on a wheelchair user and a user of a "rollator" walking frame, moving along a walkway. Observation of the sensor data in both cases showed characteristic signatures corresponding to individual "pushes". In parallel with this work, other project partners are exploring clinician and patient data requirements, as well we sensor set acceptability The initial results highlight the potential for the approach and demonstrate the need for further work to reduce and optimize the sensor set.

Preliminary study on activity monitoring using an android smart-watch.

The global trend for increasing life expectancy is resulting in aging populations in a number of countries. This brings to bear a pressure to provide effective care for the older population with increasing constraints on available resources. Providing care for and maintaining the independence of an older person in their own home is one way that this problem can be addressed. The EU Funded Unobtrusive Smart Environments for Independent Living (USEFIL) project is an assistive technology tool being developed to enhance independent living. As part of USEFIL, a wrist wearable unit (WWU) is being developed to monitor the physical activity (PA) of the user and integrate with the USEFIL system. The WWU is a novel application of an existing technology to the assisted living problem domain. It combines existing technologies and new algorithms to extract PA parameters for activity monitoring. The parameters that are extracted include: activity level, step count and worn state. The WWU, the algorithms that have been developed and a preliminary validation are presented. The results show that activity level can be successfully extracted, that worn state can be correctly identified and that step counts in walking data can be estimated within 3% error, using the controlled dataset.

Characterization of wrist-wearable activity measurement using whole body calorimetry in semi-free living conditions.

Physical activity (PA) is a significant factor in a number of health conditions and monitoring PA can play a significant role in the treatment of, or research into, these conditions. For longitudinal monitoring of PA, unobtrusive devices are often used and there is a need for the development of energy expenditure (EE) estimation techniques from single-device systems. This paper presents an experiment designed to characterize the relationship between a previously described technique, the activity score (AS) and EE obtained from whole-room indirect calorimetry. The study used 8 participants over a 24-hr period with interspersed exercise periods to observe physical movement with wearable devices and EE in 5 minute epochs. Results show that AS and EE are correlated with a Spearman's rank correlation coefficient of 0.775 with p <; 0.001.

Setting the scene: Mobile and wearable technology for managing healthcare and wellbeing.

The growing proliferation of mobile and wearable technology (MWT) offers interesting use cases when applied to health and wellness management. Current trends towards more longer term health and wellness management coupled with global challenges around the provision of healthcare to aging populations with tighter budget constraints, create rich opportunities to exploit this new technology to maintain health and wellness. This paper provides an overview of commonly available MWT and examines how it can be used in health and wellness systems. Case studies are given from two recent research projects and the issues and challenges that arise in the use of MWT are discussed. We conclude that MWT offers some key advantages in some healthcare situations, but that care must be taken to select appropriate technology for each use.

Monitoring changes in behaviour from multi-sensor systems.

Behavioural patterns are important indicators of health status in a number of conditions and changes in behaviour can often indicate a change in health status. Currently, limited behaviour monitoring is carried out using paper-based assessment techniques. As technology becomes more prevalent and low-cost, there is an increasing movement towards automated behaviour-monitoring systems. These systems typically make use of a multi-sensor environment to gather data. Large data volumes are produced in this way, which poses a significant problem in terms of extracting useful indicators. Presented is a novel method for detecting behavioural patterns and calculating a metric for quantifying behavioural change in multi-sensor environments. The data analysis method is shown and an experimental validation of the method is presented which shows that it is possible to detect the difference between weekdays and weekend days. Two participants are analysed, with different sensor configurations and test environments and in both cases, the results show that the behavioural change metric for weekdays and weekend days is significantly different at 95% confidence level, using the methods presented.

Using Empirical Mode Decomposition with Spatio-Temporal dynamics to classify single-trial Motor Imagery in BCI.

This paper introduces a new signal processing method called Spatio-Temporal Multivariate Empirical Mode Decomposition (ST-MEMD). It is a new variation of Empirical Mode Decomposition (EMD) that takes spatial and temporal information into account simultaneously rather than processing each signal source in isolation. The original and new methods were tested on single-trial electroencephalogram data with a two-class problem, in this case data using the Motor Imagery paradigm in brain-computer interfacing. However, whilst ST-MEMD retained the increase in sensitivity and specificity from adding spatial data, the new temporal data made no meaningful difference in terms of performance.

A novel method for the automatic segmentation of activity data from a wrist worn device: Preliminary results.

Activity monitoring is used in a number of fields in order to assess the physical activity of the user. Applications include health and well-being, rehabilitation and enhancing independent living. Data are often gathered from multiple accelerometers and analysis focuses on multi-parametric classification. For longer term monitoring this is unsuitable and it is desirable to develop a method for the precise analysis of activity data with respect to time. This paper presents the initial results of a novel approach to this problem which is capable of segmenting activity data collected from a single accelerometer recording naturalized activity.

Novel use of Empirical Mode Decomposition in single-trial classification of motor imagery for use in brain-computer interfaces.

This paper presents a novel method, based on multi-channel Empirical Mode Decomposition (EMD), of classifying the electroencephalogram (EEG) recordings of imagined movement by a subject within a brain-computer interfacing (BCI) framework. EMD is a technique that divides any non-linear or non-stationary signal into groups of frequency harmonics, called Intrinsic Mode Functions (IMFs). As frequency is a key component of both IMFs and the µ rhythm (8-13 Hz brain activity generated during motor imagery), IMFs are then grouped by frequency. EMD is applied to the recordings from two electrodes for each trial and the resulting IMFs are grouped according to peak-frequency band via Hierarchical Clustering Analysis (HCA). The cluster containing the frequency band of the µ rhythm (8-13 Hz) is then selected and the sum-total of the IMFs from each electrode are summed together. A simple linear classifier is then sufficient to classify the motor-imagery with 89% sensitivity from a separate test set.

Quantification of LLAEP interhemispheric symmetry by the intraclass correlation coefficient as a measure of cortical reorganization after cochlear implantation.

OBJECTIVE: Electrical stimulation by a cochlear implant (CI) induces maturation of the auditory system and reorganization of the auditory cortex in deaf children. Cortical reorganization produces an interhemispheric asymmetry in auditory evoked brain potentials associated with sound stimulation after unilateral implantation. To objectively determine the onset of this phenomenon and follow this process over time, the interhemispheric symmetry needs to be quantified. In this paper, the intraclass correlation coefficient (ICC) between mean global field powers (MGFPs) of each hemisphere is proposed to quantify long latency auditory evoked potential (LLAEP) interhemispheric symmetries as a measure of auditory cortex reorganization in CI recipients. DESIGN: An LLAEP, in response to a simple tone, was recorded in 5 juvenile unilateral CI recipients at less and at more than two years post-implantation and the ICC between MGFPs was calculated for both recordings. The cross correlation coefficients (CC) between MGFPs of each hemisphere were also calculated and compared with the ICC. RESULTS: The experience-related visually observed increases in amplitude and shape asymmetries of the LLAEP topographic map (around the LLAEP P(1) peak), were reflected in a considerable reduction of ICC values (on average 41.4%), at more than two years post-implantation surgery. In contrast, CC values only showed much smaller decreases (on average 20.0%), at more than two years post-implantation. CONCLUSIONS: The ICC is a better descriptor of symmetry than the CC because it reflects both shape and amplitude similarity between left and right LLAEP MGFPs instead of only shape similarity. The decrease in ICC values at more than two years post-implantation is likely associated with a lateralization of the auditory response as a result of cortical reorganization. Our results show that the ICC between the MGFPs for each hemisphere can be useful to objectively determine the auditory cortex reorganization process and also to evaluate the performance of cochlear implant users without the necessity to use expensive technologies such as high density EEG recordings and/or fMRI scans.

Independent component analysis for auditory evoked potentials and cochlear implant artifact estimation.

Auditory evoked potential (AEP) recordings have been analyzed through independent component analysis (ICA) in the literature; however, the performance varies depending on the ICA algorithms used. There are very few studies that concentrate on the optimum parameter selection for estimating the AEP components reliably, while also recovering the specific artifact generated with the normal functioning of a cochlear implant (CI). The objective of this research is to determine which ICA algorithm, high-order statistics (HOS)-based or second-order statistic (SOS)-based, is more plausible to remove this artifact and estimate the AEP. The optimal parameters of three such ICA algorithms for estimating the components from a database of recordings were determined, and then the estimates for the AEP and CI artifact were compared using each method. All the algorithms estimate the CI artifact reasonably well, although only one SOS algorithm is better positioned to estimate the AEP; this is primarily because it uses the temporal structure of this signal as part of the ICA process.

Space-time independent component analysis: the definitive BSS technique to use in biomedical signal processing?

Independent Component Analysis (ICA) is a very common instantiation of the Blind Source Separation (BSS) problem. In the context of biomedical signal analysis, ICA is generally applied to multi-channel recordings of physiological phenomena in order to de-noise and extract meaningful information underlying the recordings. This paper assesses the Spatio-Temporal ICA (ST-ICA) framework, which uses both spatial and temporal information derived from multi-channel time-series to extract underlying sources. In contrast, the standard implementation of the ICA algorithm generally uses only limited spatial information to inform the separation process. One of the major steps in the implementation of any ICA algorithm is the selection of relevant components from the many ICA usually returns. With ST-ICA there is a rich data-set of components exhibiting spatial as well as temporal/spectral information that could be used to identify the underlying process subspaces extracted by the ST-ICA algorithm. This paper highlights the methodology for performing ST-ICA and assesses the possible ways in which process subspace identification may take place.